laitimes

Let Musk jump "subject three", a new way to play AI video!

author:Game AI Jun

Recently, following the popular applications such as PIKA and Runaway, there is a new way to play AI video generation, swiping the screen on global social networking platforms, and the AI model Animate Anyone is online, which can make the person in the photo move by uploading pictures!

Let Musk jump "subject three", a new way to play AI video!

Last month, PIKA, known as the "video version of GPT", exploded, with a $55 million investment led by Stanford Chinese beauty Ph.D. dropping out of school to start a business, and OpenAI scientist Karpathy and Hugging Face co-founder Clem Delangue led a $55 million investment, and AI video generation is setting off a new round of craze.

Let Musk jump "subject three", a new way to play AI video!

Users only need to enter the text Prompt to change the designated area of the video to achieve the effect of one-click reloading.

Let Musk jump "subject three", a new way to play AI video!

And there is a new landing application in the field of AI video, that is, uploading pictures to make the people in the photos move, and the first thing that was targeted was last year's Guangxi "Subject 3", which was out of the circle in the global short video field and became popular on major platforms.

You only need to upload a picture of Musk, and you can let Musk jump directly to "subject three"!

Let Musk jump "subject three", a new way to play AI video!

Later, many netizens made videos of Apple CEO Cook, Meta CEO Xiaozha and others, and the video action looks relatively smooth, and even fake!

Let Musk jump "subject three", a new way to play AI video!

This time, the AI video is mainly based on the AI model Animate Anyone, and the Miao Ya camera that swiped the screen in the circle of friends last time is also from the hands of the Ali team.

Let Musk jump "subject three", a new way to play AI video!

Animate Anyone wants to be able to animate videos for anyone, giving a picture of a character with a distinct character pose to generate controllable images and video animations, with the same diffusion models at its core.

Let Musk jump "subject three", a new way to play AI video!

The project designed the ReferenceNet algorithm to merge detailed features through Spatial-Attention. Ensure that the action sequences generated by the picture can be coherent, and these action sequences can be dancing, walking, etc.

Let Musk jump "subject three", a new way to play AI video!

Paired with Outfit Anyone, you can design and change outfits for any character and generate animations for a wide range of applications in the future!

Let Musk jump "subject three", a new way to play AI video!

Read on